Nyhed
Johannes Bjerva and the hidden danger of AI
Lagt online: 11.08.2025

Nyhed
Johannes Bjerva and the hidden danger of AI
Lagt online: 11.08.2025

Portrait - professor Johannes Bjerva
Nyhed
Lagt online: 11.08.2025
Nyhed
Lagt online: 11.08.2025
By Peter Witten, AAU Communication and Public Affairs
How did you become interested in researching what you specialize in today?
"I’ve always been very interested in language, so after high school I chose to study Japanese for a year before starting a degree in linguistics. Halfway through my studies, I discovered an entire research field focused on the automatic analysis of language – Natural Language Processing (NLP), more specifically the branch called Computational Linguistics. I thought it sounded like a very exciting opportunity, even though research in that area was quite limited at the time. With new methods like large language models (LLMs), the field has really taken off. To illustrate: I just attended our biggest conference, which had over 6,000 participants and around 3,000 papers. That’s more than ten times the number when I was a PhD student eight years ago," says professor Johannes Bjerva.
What was your path from Norway to Aalborg University - and why AAU?
"I grew up in the countryside in Norway and later in the small town of Hamar. After high school at Hamar Cathedral School, I wanted to move somewhere bigger and chose Stockholm. As a bachelor’s and master’s student there, I saw how exciting it seemed to be a PhD student. I decided to pursue that path, which led to a PhD position in Groningen, the Netherlands, where I lived for about four years. Ending up in Denmark was more or less coincidental. Right after my PhD, I was offered a postdoc position at DIKU (Department of Computer Science, University of Copenhagen), and I enjoyed living in Copenhagen. In 2020, I saw a very exciting job posting for a tenure-track assistant professorship at AAU in Copenhagen. I read that it involved being a central part of launching a new program (Software) and building an entirely new section."
Johannes Bjerva continues: "There was no NLP research at AAU in 2020, so for me it was also a strategic choice to try to build something new - both in education and as a research group. Today, it’s a rapidly growing group, supported by the generous research grants I’ve received in recent years. In 2021, I received a Semper Ardens: Accelerate grant from the Carlsberg Foundation, which has been absolutely crucial for my career. Based on that project, we are now established as a high-quality research environment in NLP, with impact both on society and the research community. In my view, it was this first grant from the Carlsberg Foundation that enabled me to develop the good ideas that, in the past six months, have led to my DFF: Sapere Aude, NNF: Data Science Investigator, and also the AAU-funded AI:X lab. All in all, we expect to have around 10 PhD students in NLP by summer 2026."
What does it mean for your research that it takes place at AAU?
"There are several concrete advantages to being at AAU. Among other things, PBL (Problem-Based Learning) gives me and the other researchers in the group a very close connection to the students. This means, for example, that we can work with smaller research-relevant building blocks in group projects in the Software program, and that we have plenty of opportunities to identify top students and recruit the very best for PhD positions. In fact, one of the PhD students starting in our AI:X Lab is from the very first Software cohort at AAU-Copenhagen. The most important thing for me is probably the freedom I’ve had to pursue the path I believe makes the most sense. I’ve really experienced strong support for my research, career development, and for establishing a new group - from colleagues, department leadership, and faculty level."
Why is it so important to research what you’re working on now? (TRUST)
"Language models (LLMs) and generative AI are already widespread across society, and this will continue as we find relevant use cases and the models improve. However, many people aren’t aware that such language models can be influenced by malicious actors to introduce more or less subtle bias. TRUST is an essential building block to secure such models and enable us to detect if a language model has been externally influenced."
What are the societal perspectives of your research?
"There are two main perspectives. One is to communicate that the issue exists. Many people know that a language model can 'hallucinate' or produce false information. But I don’t think many know that a language model can be subtly influenced to generate content that follows a specific political agenda. Imagine if everyone’s favorite chatbot was subtly trying to make all EU citizens a little more far-right every day, without ordinary people noticing the rhetorical techniques it uses. The second is to develop the technology so that the models are actually more secure. If we succeed, we’ll be able to reduce how often a language model is externally influenced."
How can your research benefit others in terms of research?
"TRUST is basic research in my field, so the goal is primarily to create advances that can influence how future research is conducted and serve as a foundation for future research and technology. Among other things, the project has a hypothesis that using linguistic knowledge can give us much more effective defense mechanisms. The two PhD students and the postdoc working on TRUST will have the opportunity to work on something that’s becoming very hot in NLP. They’ll be among the first to earn a PhD and gain research experience in this area. They’ll have a unique and highly sought-after skill set when the project is completed. Very large international companies are increasingly interested in this new research area."
A bit more about yourself?
"I just turned 35. Next year, I’ll have lived abroad for half of my life so far. A couple of years ago, my wife and I settled in Farum, and we’ve just become parents to our third child. It’s pretty hectic with three small children aged 4, 2, and 0, so it can be challenging to balance work and family life - especially since being a researcher often involves a lot of travel, perhaps even more so when working at a university with multiple campuses. In my free time, I used to be quite active in long-distance training, mainly running and triathlon, and I completed Ironman Copenhagen in 2022. Training has unfortunately been on pause since then, but I’m working on prioritizing it again."
Facts